The open-source AI ecosystem has welcomed another major update. Tencent Hunyuan officially announced on November 18, 2026, that it is open-sourcing the Hy4flash model, bringing developers and users a brand-new large model option focused on ultimate cost-effectiveness.
The full family matrix makes its debut, with multi-channel support for convenient deployment
The open-source versions released by Tencent Hunyuan this time cover multiple specifications, specifically including three versions: Hy4flash, Hy4flash-FP8, and Hy4flash Base.
To help global developers quickly get started and conduct secondary development, Tencent has simultaneously opened up multiple mainstream code hosting and model community platforms. Developers can currently download and deploy locally through channels such as GitHub, Hugging Face, ModelScope, and GitCode. The community also provides corresponding GGUF format support and detailed open-source license documentation. In addition, Tencent Hunyuan officials also revealed that they are accelerating the open-sourcing of more Hy4 series models in the future.

Core architecture and performance: A perfect balance of large capacity and high efficiency
As the most cost-effective Hy model to date, Hy4flash shows many highlights in parameter scale and actual performance:
Parameters and activation: The model has a total parameter count of 224B, including 64B Engram parameters; in actual operation, only 12B parameters are activated per token, balancing strong performance with extremely high computational efficiency.








